Barriers, Lock-in & Strategic Behavior

CHAPTER 11

Barriers, Lock-in & Strategic Behavior

Why concentration persists and how incumbents defend position

Joe Bain's classic taxonomy identifies three sources of barriers to entry: absolute cost advantages, economies of scale, and product differentiation. The AI supply chain exhibits all three — plus a fourth that Bain could not have anticipated: ecosystem lock-in through proprietary software platforms.

💰

Irreversible Sunk Costs

A chip fab costs $20-40 billion and takes 4-5 years to build. An EUV machine costs $400M. These investments cannot be recovered if entry fails, creating asymmetric risk between incumbents and entrants.

📈

Minimum Efficient Scale

Cloud providers need millions of customers to justify infrastructure costs. Foundries need consistent utilization to achieve competitive yields. Below MES, unit costs are prohibitively high.

📉

Learning Curves

TSMC's yield advantage comes from cumulative production experience — billions of chips manufactured. Wright's Law: costs fall 20-30% with each doubling of cumulative output.

🔗

Network Effects

CUDA's value increases with each developer who learns it and each application optimized for it. Two-sided platform economics: hardware attracts software which attracts hardware.

Once established, dominant firms face a strategic choice that economists have debated since Schumpeter and Arrow. Do high profits fund continued innovation ("competition for the market"), or do incumbents shift resources from creating value to defending position?

Schumpeter / Competition for the Market

High margins fund frontier R&D

Nvidia's >70% gross margins fund $10B+ annual R&D. TSMC reinvests ~$30B/year in capacity. ASML spent 30 years developing EUV. Without the prospect of supernormal returns, these investments would not occur.

Arrow (1962): Monopolist has less incentive to innovate — it would only replace itself

Entrenchment / Position Protection

Dominance enables exclusionary tactics

Loyalty rebates that make rivals' contestable share unprofitable. Killer acquisitions of potential competitors. Tying and bundling across adjacent markets. Supply allocation that favors internal divisions or preferred partners.

Gilbert & Newbery (1982): Incumbent invests to preempt, not to innovate

$20B+

TSMC+ASML+Nvidia R&D (2024)

405%

Nvidia revenue growth 2023-24

3 yrs

GPU shortage duration

Supply shortages reveal how market power operates in practice. GPU demand has exceeded supply for three consecutive years. When supply is constrained, dominant suppliers can allocate strategically — prioritizing internal projects, favoring partners, or conditioning supply on other commercial terms.

Antitrust Case Studies

DRAM Price-Fixing Cartel (1998–2002)

COLLUSION

As the DRAM market matured from innovation-driven competition to commodity production, Samsung, Hynix, Infineon, Elpida, and Micron formed a price-fixing cartel. Executives coordinated prices through direct communication, exploiting the oligopolistic market structure.

$645M

US fines

EUR 331M

EU fines

Immunity

Micron (whistleblower)

Intel Loyalty Rebates (EC 2009)

EXCLUSION

Intel offered Dell, HP, Lenovo, and NEC conditional rebates for purchasing exclusively or near-exclusively Intel processors. The theory of harm: rebates made the "contestable share" — the portion of demand a rival could realistically win — so low-margin that AMD could not profitably compete, even with superior technology.

EUR 1.06B

EC fine

10 yrs

Duration of conduct

2024

Final appeal rejected

What starts as innovation-driven dominance may become entrenched market power once a firm shifts from creating value to defending position. The challenge for competition policy is distinguishing between the two.

Switching Costs & Lock-in Mechanisms

Contractual

Multi-year volume commitments

Exclusivity clauses

Cloud egress fees

Most-favored-nation clauses

Technical

Proprietary interfaces

Data format lock-in

Firmware dependencies

Retraining requirements

Ecosystem

CUDA: 4M+ developers

3,000+ optimized applications

Training materials & courses

Complementary investments

Key Relationships

Vertical Foreclosure

Control of one layer enables restricting competition in adjacent layers. A cloud provider that also makes chips can favor its own hardware; a chipmaker that owns models can prioritize its own inference.

Killer Acquisitions

Acquiring potential competitors before they reach scale. In concentrated markets, the incumbent often values a startup more than financial buyers — specifically to prevent competition.

Tying & Bundling

Conditioning access to a bottleneck product on purchasing a competitive product. Classic theory: leverage monopoly in A to foreclose competition in B.